ask

CLI for querying LLMs from the terminal. Supports Anthropic (Claude), OpenAI, and Google (Gemini).

Talks to each provider's REST API directly over net/http — no vendor SDKs, so the only dependency is a TOML parser.

Install

go install github.com/leinonen/ask@latest

Or build from source:

go build -o ask .

Usage

ask "what is the capital of France?"
echo "func foo() {}" | ask "explain this"
cat file.txt | ask "summarize"
ask --shell "list files sorted by size"
ask --code "fibonacci in python"
ask --caveman "explain what a mutex is"

Flags

Flag	Short	Description
--shell	-s	Generate a shell command (prompts to execute)
--code	-c	Output code only, no markdown fences
--caveman	-V	Terse caveman output (~75% fewer tokens)
--system TEXT	-S	Custom system prompt
--model TEXT	-m	Model override
--provider TEXT	-p	Provider override (anthropic, openai, or gemini)

Config

On first run, a setup wizard prompts for provider, API key, and default model. Config is saved to ~/.config/ask/config.toml.

To enable caveman mode by default, add caveman = true to your config:

provider = "anthropic"
model    = "claude-opus-4-7"
caveman  = true

You can also set keys via environment variables:

export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...
export GEMINI_API_KEY=...   # GOOGLE_API_KEY also works

Gemini

Select Google (Gemini) in the setup wizard, or set it in your config:

provider = "gemini"
model    = "gemini-2.5-flash"

[gemini]
api_key = "..."

Get a key from Google AI Studio. Available models include gemini-2.5-flash, gemini-2.5-pro, gemini-2.5-flash-lite, and gemini-2.0-flash.

OpenAI-compatible providers

Any OpenAI-compatible service (Ollama, LiteLLM, Azure OpenAI, local models, etc.) works by setting base_url in the [openai] section:

provider = "openai"
model    = "llama3.2"

[openai]
api_key  = ""
base_url = "http://localhost:11434/v1"

The setup wizard also prompts for a base URL when OpenAI is selected. Leave it blank to use api.openai.com.